Tracking landslide displacements by multi-temporal DTMs

Tracking landslide displacements by multi-temporal DTMs: a
combined aerial stereophotogrammetric and LIDAR approach in
western Belgium
Dewitte O.a, *, Jasselette J.-C.b, Cornet Y.c, Van Den Eeckhaut M.d, e, Collignon A.b,
Poesen J.d, Demoulin A.a, f
a
Department of Geography, Unit of Physical Geography and Quaternary, University of
Liège, Allée du 6 Août 2, 4000 Liège, Belgium
b
Ministry of Equipment and Transport of the Walloon Region, Boulevard du Nord 8,
B-5000 Namur, Belgium
c
Department of Geography, Unit of Geomatics, University of Liège, Allée du 6 Août
17, B-4000 Liège, Belgium
d
Physical and Regional Geography Research Group, K.U. Leuven, Celestijnenlaan 200E, 3001 Heverlee, Belgium
e
Post-doctoral fellowship, Fund for Scientific Research −Flanders, Belgium
f
Belgian National Fund for Scientific Research, Belgium
* Corresponding author. Department of Geography, Unit of Physical Geography and
Quaternary, University of Liège, Allée du 6 Août 2, 4000 Liège, Belgium.
Tel.: +32 43 665255; fax: +32 43 665722.
E-mail address: [email protected] (O. Dewitte)
1
1
Abstract
2
3
The study of small and/or slow reactivations of landslides requires describing their
4
displacements over decades, which may be done with accurate multi-temporal digital
5
terrain models (DTMs). We applied aerial stereophotogrammetry to build the historical
6
topographies of old deep-seated landslides close to Oudenaarde in the Flemish
7
Ardennes (West Belgium) at different dates. Three precise aerotriangulations (1996,
8
1973, 1952) were carried out. After capturing the ground data manually from the
9
stereomodels, 2 m-resolution DTMs were interpolated by kriging, with a final accuracy
10
ranging between ~45 cm and ~65 cm. Another DTM was interpolated with an accuracy
11
of ~30 cm from airborne LIDAR data acquired in 2002. Differential DTMs were
12
produced to identify vertical and horizontal ground displacements over the 1952-2002
13
period. We describe here the kinematics of a particularly active landslide with a well-
14
documented recent activity. Until the first half of the 90s, little activity of the landslide
15
was detected. In February 1995 a reactivation event caused vertical displacements of up
16
to -7 m along the main scarp and up to +4 m in the accumulation zone. Horizontal
17
movements of 4 to 10 m are also inferred. These topographic changes correspond to
18
reactivated slip along the rotational basal shear surface. In the same time, the main scarp
19
retreated by up to 20 m. The reactivation, favoured by several anthropogenic factors
20
(e.g. loading, impeded drainage), was triggered by intense rainfall. Between 1996 and
21
2002, the observed displacements correspond to limited scarp retreat (≤ 4m) and
22
compaction of the slipped mass, partly enhanced by artificial drainage.
23
2
24
Keywords: Landslide kinematics; Photogrammetry; LIDAR; differential DTM;
25
Landslide reactivation; Belgium.
26
27
28
1. Introduction
29
30
The understanding of landslide mechanisms is greatly facilitated when information on
31
their horizontal and vertical displacements is available. Geodetic techniques, recently in
32
particular GPS, are commonly used to monitor ground motion with a high accuracy
33
(Malet et al. 2002; Coe et al., 2003; Squarzoni et al., 2005; Brückl et al., 2006;
34
Demoulin, 2006). However, they require access to the site and may be time-consuming
35
if the information has to be densified, spatially and/or temporally. Moreover such
36
studies generally encompass a time-interval of a few years at maximum, thus providing
37
poor information on longer-term landslide kinematics.
38
39
According to the scale, accuracy and resolution needed, several techniques of remote
40
sensing are available to build digital terrain models (DTMs) of landslides (Mantovani et
41
al., 1996; Metternicht et al., 2005). In the case of the Flemish Ardennes (Fig. 1), the
42
ground displacements are rather small (in the order of few meters) and slow, and their
43
reliable description requires a final DTM accuracy better than 1 m in the three
44
directions. Several techniques such as GB_InSAR (Tarchi et al., 2003; Antonello et al.,
45
2004), LIDAR (McKean and Roering, 2004; Chen et al., 2006; Glenn et al., 2006), GPS
46
(Higgitt and Warburton, 1999; Nico et al., 2005; Demoulin, 2006), and aerial
47
photogrammetry (Chandler, 1999; Hancock and Willgoose, 2001) are capable of
3
48
supplying such accurate 3D topographic data. However, only the
49
stereophotogrammetric analysis of aerial photographs can cover several decades of
50
observation (Hapke, 2005).
51
52
The main objective of this research is to assess how reliable landslide small
53
displacements can be measured from multi-temporal DTMs produced by aerial
54
stereophotogrammetry. For this purpose, we used classical aerial photographs at scales
55
ranging between 1:18 500 – 25 000 (Table 1), which required careful measurements at
56
each step of the DTM production in order to reach a submeter accuracy (Kraus and
57
Waldhäusl, 1994; Kasser and Egels, 2001). A detailed analysis of the precision,
58
accuracy and reliability associated with each step of the DTM extraction procedure is
59
therefore given. We present first the photogrammetric procedure applied to the
60
photographs. Details on the subsequent DTM construction and on a LIDAR-derived
61
DTM provided by the Flemish Government (DEM of Flanders, 2007) are also given.
62
We then show how vertical and horizontal landslide motions can be inferred from DTM
63
subtraction, focusing on one particular landslide, whose activity is well documented
64
over the last 50 years.
65
66
2. Study area
67
68
In West Belgium, Van Den Eeckhaut et al. (2005) mapped more than 150 large deep-
69
seated landslides within the loose Tertiary sediments of the hilly area of the Flemish
70
Ardennes (Fig. 1). All landslides predate 1900 AD and might have been initiated under
71
periglacial conditions, possibly in response to a seismic trigger associated with a period
4
72
of heavy rainfall (Van Den Eeckhaut, 2006; Van Den Eeckhaut et al., 2007b).
73
According to the Keefer’s relation between maximum distance of landslides from fault
74
and Richter magnitude ML (Keefer, 1984), Van Den Eeckhaut (2006) estimated
75
magnitudes between 5 and 6 to trigger landslides in the Flemish Ardennes. However,
76
the biggest earthquake recorded in Belgium during the 20th century occurred within the
77
Flemish Ardennes in June 1938 and, despite a magnitude ML of 5.6, no landslide was
78
reported.
79
80
Fig. 1
81
82
This research focuses on two hills affected by 13 landslides close to the town of
83
Oudenaarde and situated along the river Scheldt at altitudes between 75 and 80 m a.s.l.
84
(Fig. 1). The landslides are carved in an alternation of Eocene subhorizontal clayey sand
85
and clay layers on which a perched water table can expand (Jacobs et al., 1999). The 13
86
landslides are in contact with the Aalbeke Member of the Kortrijk Formation, which
87
consists of 10-m-thick homogeneous blue massive clays and has been recognized as the
88
layer most sensitive to landsliding (Fig. 1). The mean landslide size is ~6 ha and their
89
main scarp, with a height of ~8 m, makes the abrupt fringe of the plateaus (Dewitte and
90
Demoulin, 2005). They developed on slopes of 13-20 %, preferentially oriented to the
91
W and N. The mean value of the depth of the surface of rupture has been estimated in
92
the range 30 - 70 m, implying displaced volumes between 670 000 m³ and 1 500 000
93
m³.
94
5
95
3. Photogrammetric operations
96
97
Aerial photographs are frequently used in digital photogrammetry to extract landslide
98
relief (e.g. Oka, 1998; Weber and Hermann, 2000; Kääb, 2002; Baldi et al., 2005;
99
Casson et al., 2005; Hapke, 2005; Lantuit and Pollard, 2005). We selected only the
100
aerial covers supplied by the National Geographical Institute (NGI) of Belgium that
101
were taken at the beginning of the spring (when the trees are leaveless) in order to
102
obtain a better restitution of the landslide areas under forest. We thus used the
103
photograph negatives of the 1952, 1973 and 1996 covers, allowing the investigation of
104
the landslide kinematics over a 44-year period (Table 1).
105
106
Table 1
107
108
The image processing (Fig. 2) was performed using the Leica Helava Digital
109
Photogrammetric Workstation DPW 300 with the software SOCET SET v4.3.1b (LH
110
Systems, 1999). The photograph negatives were scanned with the LH Systems DSW
111
300 Digital Scanning Workstation with a pixel resolution of 12.5 μm that corresponds
112
to a ground resolution of approximately 20 to 30 cm (Table 2).
113
114
Fig. 2
115
116
3.1. Interior orientation
117
6
118
The interior orientation was carried out during the scanning operation (Fig. 2) by using
119
the parameters of the camera established by the geometric calibration (Table 1). The
120
radial distortion was only considered for the 1996 negatives for which a camera
121
calibration certificate was provided. For the covers of 1973 and 1952, the radial
122
distortion was not implemented and dummy fiducial marks were used for the 1952
123
negatives.
124
125
Table 2.
126
127
The final accuracies of the interior orientation are shown in Table 2. With accuracy
128
values better than 10 cm, we may consider that the root mean square error (RMSE) of
129
this orientation will not act significantly upon the final accuracy of the stereomodels.
130
The larger error of 1952 is probably due to the use of less reliable fiducial marks.
131
132
3.2. Aerotriangulation
133
134
The aerotriangulation of the digital photos was performed by using an iterative least-
135
square bundle block adjustment allowing simultaneously the relative and the absolute
136
orientations of the photographs in the block (Kraus and Waldhäusl, 1994; Mikhail et al.,
137
2001). The absolute orientation of the block requires the use of ground control points
138
and the block adjustment accuracy is evaluated using check points with known ground
139
coordinates (Fig. 2) which are not used as control in the solution. Moreover, control
140
points and check points must be accurate enough to reflect the actual state of the ground
141
surface.
7
142
143
A total of 26 photo-identifiable ground control points and check points acquired by GPS
144
in February 2002 were used for the adjustments. The measurements were carried out by
145
differential GPS in rapid static mode involving baselines of 1 to 6 km length. The
146
uncertainty on the baseline components (N, E, and Up) is ~2-3 cm. Absolute positioning
147
of the control and check points relies on the inclusion of five NGI points of known
148
coordinates within the GPS network. The final uncertainty on the coordinates of the
149
control and check points do not exceed 10 cm (Dewitte, 2006), thus remaining far
150
below that of the block adjustment. Note that the GPS points were measured at places
151
very easy to locate on the 1996 photographs (electric poles, fences, crossroads, etc…).
152
153
Three aerotriangulations were built, one for each covering date (Table 3). Each bundle
154
block adjustment requires overabundant data to attribute X, Y and Z ground coordinates
155
to each tie point (Kraus and Waldhäusl, 1994; Mikhail et al., 2001). As stressed by LH
156
Systems (1999), the number of required points to obtain a logic solution depends on the
157
number of image parameters adjusted, the number of images each point appears on, and
158
the availability of the ground points (Table 3). For each image to adjust, 6 orientation
159
parameters need to be computed. Each tie point contributes 3 unknowns (ground X, Y,
160
Z) but provides 2 knowns (line ξ and sample η) for each image it is measured on. A
161
ground control point brings more knowns than a tie point since some or all its ground
162
coordinates are known.
163
164
Table 3
165
8
166
The first stereomodel was adjusted with the three photographs of 1996 and 16 ground
167
control points (Table 3). Each of these points was measured on two images (4 knowns)
168
and its 3 ground coordinates (XYZ) were provided too (3 knowns). The adjustment of
169
the 1996 stereomodel was performed in ideal conditions since all the parameters of the
170
interior orientation were provided and the image quality was very high. The tie points
171
were automatically measured before being interactively reviewed and corrected whereas
172
the image positioning of the control points acquired by GPS (Fig. 2) was made
173
manually in stereoscopy. Moreover, the control points were easy to locate.
174
175
However, the positioning of control points on historical photographs is far more
176
difficult (Hapke, 2005). To avoid inaccuracies associated with the positioning of these
177
points, the two stereomodels of 1973 and 1952 were rectified relative to the 1996
178
stereomodel (Hapke, 2005). For each new aerotriangulation, the images were placed in
179
a second strip associated with the 1996 model whose adjusted orientation parameters
180
had been blocked. The two other stereomodels were thus constructed by including the
181
orientation parameters of 1996. Therefore, only tie points whose absolute XYZ
182
coordinates are computed in the model of 1996 were necessary for the construction of
183
the 1973 and 1952 aerotriangulations (Table 3). The tie points common to 1996 and
184
either 1973 or 1952 acted here as control for the absolute orientation. All the points
185
used in the three aerotriangulations were chosen outside the landslides in order to rely
186
on stable places only (Berthier et al., 2005).
187
188
The measurement redundancy (LH Systems, 1999; Mikhail et al., 2001) is positive for
189
the three bundle block adjustments (Table 3) and the required condition of overabundant
9
190
data is therefore fulfilled. Moreover, the percent redundancy, defined as the ratio
191
between redundancy and total number of ground points, is higher than the 50% value
192
recommended by LH Systems (1999), giving to the solution a more consistent and
193
meaningful residual RMSE value.
194
195
3.3. Evaluation of the aerotriangulations
196
197
A qualitative and a statistical evaluation of the aerotriangulations were carried out to
198
ascertain that the block adjustments meet our requirements and are valid.
199
200
The qualitative evaluation consisted in displaying the residuals of the aerotriangulations
201
(Figs. 3 and 4) in order to understand broad trends and to catch bad inputs. The ground
202
points are distributed uniformly in the three stereomodels and the image residuals are
203
smallest in the 1996 model (Fig. 3). The image residuals of the 1973 and 1952
204
adjustments point in random directions and have comparable sizes around the model
205
centres where the landslides are located (Fig. 3). On the contrary, the size of the 1996
206
residuals often varies considerably from one point to another. This might result from too
207
tight constraints applied to the ground control points, which did not allow them to adjust
208
enough (Dewitte, 2006). As the residuals were very close to the maximum precision of
209
0.5 pixel announced by LH Systems (1999) (Table 4), the small E-W trend observed in
210
the 1996 model was neglected.
211
212
Fig. 3
213
10
214
Fig. 4
215
216
The check points we used were placed close to the two investigated landslide areas (Fig.
217
4). No deformation appears in the block of 1996 and 1973, whereas the SW part of the
218
1952 model is slightly higher. The errors are randomly distributed and of comparable
219
sizes.
220
221
In brief, the qualitative evaluation shows that the three aerotriangulations are reliable
222
and respond to the requirements of this study. Like the image residuals (Fig. 3), the
223
XYZ residuals (Fig. 4) of 1996 are the lowest and those of 1952 the biggest.
224
225
The statistical evaluation of the adjustments was performed by computing the precision
226
and the accuracy of the measurements. The fact that the mean of the image residuals is
227
very close to zero in the three models (Table 4) suggests that we are in the “ideal case”
228
where the residuals may be treated as normally distributed random variables (Buiten and
229
van Putten, 1997; Mikhail et al., 2001). The precision of the solution is given by the
230
standard deviation of the image residuals of the ground points. For the 1996
231
aerotriangulation, the total standard deviation lower than 1 pixel satisfies the quality
232
criterion given by LH Systems (1999) (Table 4). The somewhat lower precision of the
233
two other solutions is not due to bad measurements of ground points. Rather, the
234
measurement of the tie points of 1973 and 1952 using their coordinates in the 1996
235
images implied a combination of the errors of both models. Moreover, the resolution of
236
the 1952 images is coarser (Table 2).
237
11
238
Table 4
239
240
We estimated the accuracy of the solutions with 8 to 10 check points measured by GPS
241
(Table 4). The evaluation was based on the RMSE of the check point differences
242
between the computed coordinates generated for each block and the ground truth (GPS
243
values). Since it was more than one order of magnitude better than the accuracy
244
required for the final DTMs, the accuracy of the GPS measurements was not considered
245
in the RMSE computations. Although the number of check points is not very high, they
246
are located close to the landslide areas and we thus consider that they yield reliable
247
results. If the errors are assumed to be normally distributed and independent, they can
248
be used as standard deviation and the combined variance of independent variables is
249
given by the sum of the individual variables variances (Borradaile, 2003). The
250
horizontal RMSE values are similar to the vertical RMSE values and the worst values
251
are observed in the 1952 aerotriangulation (Table 4). The global RMSE values range
252
between ~22 cm and ~55 cm for the three stereomodels and are in agreement with the
253
theoretical values proposed by Kraus and Waldhäusl (1994) and by Kasser and Egels
254
(2001).
255
256
With respect to our goal, which is to locate vertical ground movements > 1 m, the
257
qualitative and statistical evaluations have shown that the three stereomodels are of
258
sufficient reliability, precision, and accuracy.
259
260
3.4. Stereoscopic data capture
261
12
262
The automatic terrain extraction methods provided with SOCET SET were not able to
263
generate the DTMs accurately in many places, especially in the forested areas
264
(Chandler, 1999; Fabris and Pesci, 2005). The stereoscopic data capture of spot heights
265
(ground points) and breaklines (scarps, roads, water bodies) on the landscape surface
266
was therefore carried out visually. The spot heights were extracted approximately every
267
10 m, but with a higher density in more contrasted topography, in particular within the
268
landslides. In total, six areas were surveyed, i.e. the two investigated hills at three
269
different dates (Fig. 1).
270
271
4. DTM construction
272
273
The captured elevation data were interpolated by ordinary kriging with the SURFER 8.0
274
software. Ordinary kriging is one of the most widely-used method in geostatistics and it
275
has proved very effective in the interpolation of topographic data for the generation of
276
DTMs (e.g. Siska et al., 2005; Chaplot et al., 2006). Through the use of
277
semivariograms, the kriging method considers the spatial autocorrelation in the
278
topographic data (Goovaerts, 1997). This interpolation method also allows the insertion
279
of breaklines which depict abrupt changes in elevation and increase the accuracy of the
280
DTMs.
281
282
All six DTMs were interpolated by using an omnidirectional semivariogram adjusted
283
with a power model (Goovaerts, 1997). Taking into account the data density, their
284
spatial autocorrelation shown by the semivariograms, the precision and the accuracy of
285
the stereomodels, the DTM grids were generated at a 2-m resolution. The vertical
13
286
RMSE of the kriging interpolations obtained by cross-validation are ~40 cm (Table 5).
287
The slightly higher error value of the 1996 DTM of the Leupegem hill with respect to
288
that of the Rotelenberg is the consequence of its higher topographic variability due to
289
the recent reactivation of landslide 1 (Fig. 1).
290
291
Table 5
292
293
Assuming that the kriging and aerotriangulation errors are normally distributed and
294
independent, the total vertical errors of the DTMs range from 0.4 to 0.5 m (Table 5) and
295
the total 3D errors range between ~0.45 and ~0.65 m, which corresponds to values far
296
below our required 1 m accuracy for the final DTM.
297
298
At the end of 2002, airborne LIDAR (Light Detection and Ranging) data were acquired
299
over the study area by the Flemish Government with a point density of 1 point per 4 m².
300
After correcting the dataset for canopy and building returns (DEM of Flanders, 2007), a
301
database with a point density of at least 1 per 20 m² was provided with an accuracy
302
varying according to the land cover type (vertical RMSE of at worst 0.20 m under forest
303
or better). Using a TIN interpolation, we then derived from the LIDAR data a DTM
304
with a resolution of 2 m. The maximum vertical error of the interpolation being 0.2 m
305
(Dewitte, 2006), the total RMSE in Z for the 2002 DTM equals ~0.30 m (Table 5).
306
307
5. Results: 1952-2002 evolution of a typical landslide
308
14
309
As an example, we detail the movements recorded for the landslide 1 of the Leupegem
310
hill (Fig. 1). According to size and lithology, this landslide is representative not only of
311
those of the Leupegem and Rotelenberg hills but also of those of the whole Flemish
312
Ardennes (Dewitte and Demoulin, 2005). Moreover, its reactivation in February 1995
313
caused comparatively large ground displacements, and eye-witness reports of its
314
evolution over the past 50 years are available (Van Den Eeckhaut et al., 2007a),
315
allowing some qualitative validation of the measured displacements. This landslide
316
extends over an area of ~9 ha and has a total length of 370 m and a maximum width of
317
260 m. The maximum height of its main scarp is 9 m. Through morphometric
318
measurements, Dewitte and Demoulin (2005) estimated the depth of the rupture surface
319
and volume before movement. Supposing a single rotational earth slide, they have
320
computed a pre-landslide volume of 1 250 000 m³ for a depth of the surface of rupture
321
of 40 m.
322
323
5.1. Vertical displacements
324
325
A simple DTM subtraction yields a direct estimate of the vertical displacement of each
326
landslide pixel during the considered time interval (e.g. Oka, 1998; Weber and
327
Herrmann, 2000; Gentili et al., 2002; Kääb, 2002; van Westen and Getahum, 2003;
328
Baldi et al., 2005; Casson et al., 2005; Hapke, 2005; Lantuit and Pollard, 2005; Brückl
329
et al., 2006; Chen et al., 2006; Demoulin 2006). DTM subtraction implies that
330
individual DTM errors can be combined as independent random variables. With the
331
assumption of normal distribution, this combined error can be used as standard
332
deviation (Mikhail et al., 2001). Note that, since the horizontal error is a two-
15
333
dimensional error, the factor for 68.3% confidence is 1.55 and 2.45 for 95% confidence
334
(Mikhail et al., 2001). The confidence intervals associated with the ground
335
displacements of the landslides are presented in Table 6. At a 68.3% confidence level,
336
we are thus able to detect vertical ground displacements of ~0.70 m or larger.
337
338
Table 6
339
340
We present in figure 5 the vertical movements identified within this recently reactivated
341
landslide. These maps are produced with a confidence interval of 68.3%, which means
342
that only differences in elevation larger than 1 σ uncertainty are considered. For the
343
sake of consistency, these maps are all shown with a same confidence value of ± 0.75
344
m.
345
346
Fig. 5
347
348
Between 1952 and 1973 (Fig. 5A), the topographical changes were very limited. The
349
main scarp remained unchanged. The small collapses observed at the head of the slide
350
might have resulted from destabilization and tension cracks in response to high
351
groundwater levels at the contact with the Aalbeke clays as it is currently observed in
352
other landslides of the Flemish Ardennes. The role of runoff is poorly known. No data
353
and field work are available and one even ignores whether it induced a net loss of
354
volume within the landslide, or rather brought material from upslope of the main scarp.
355
Runoff was however probably of little importance due to the presence of a forested band
356
upslope of the landslide head during all this period (Dewitte, 2006). The elongated zone
16
357
of uplift observed along the Hekkebrug road north of house 2 was due to improvement
358
works that consisted in adding layers of stone bricks and asphalt to increase road
359
accessibility (Van Den Eeckhaut et al., 2007a).
360
361
By contrast, significant movements occurred during the 1973-1996 period (Fig. 5B),
362
revealing a typical rotational pattern, with a subsiding zone at the head, an intermediate
363
area without vertical displacement, and an uplifted zone of accumulation towards the
364
foot of the landslide (Dikau et al., 1996). Immediately downslope of the main scarp,
365
large values of apparent collapse, corresponding to the height of the scarp (~7 m),
366
actually indicate a scarp retreat. The amount of retreat can be approximated by the
367
width of the band of collapsed terrain and reaches locally 15 m. The ground changes
368
identified around the ponds near house 2 (Figs. 5B and C) are related to the construction
369
of a garden in 1993. Note that the areas unaffected by the landslide, upslope of the main
370
scarp, show no significant height difference between the two models, thus validating the
371
comparison.
372
373
A study by Van Den Eeckhaut et al. (2007a) pointed out that the movements of the
374
1973-1996 period were mainly due to the reactivation event that occurred in February
375
1995. This reactivation was triggered by the heavy rainfall of the winter 1994-1995.
376
Their effect was enhanced by increased surface runoff resulting from the development
377
of cultivated areas that expose bare soils in winter upstream of the main scarp, with the
378
consequence of an increase of the quantity of water directed to the landslide. Moreover,
379
the artificial loads aimed at the improvement of the Hekkebrug road and the digging of
380
several ponds close to house 2, and the absence of well-maintained drainage ditches
17
381
after the cessation of the agricultural activities in the landslide in the 50s contributed
382
also to reactivation.
383
384
After 1996 (Fig. 5C), a decrease of the landslide activity was observed. It is
385
characterized by the subsidence of large areas within the landslide, due mainly to
386
compaction in response notably to the digging of two 2-m-deep trenches across the
387
affected area in 2000 (Figs 5C and E). Starting from the base of the main scarp, these
388
trenches of ~250 m in length (total volume of both trenches = 6 000 m³) drain the water
389
springing from the aquifer perched above the Aalbeke clays, and the runoff water.
390
Subsequently, little displacement has been observed in the zone of accumulation since
391
2001 (Van Den Eeckhaut et al., 2007a). The main scarp however was still active during
392
this period as pictured in Fig. 5F. This soil slice of ~25 m length and 4 m width
393
collapsed in March 2002. Soil material was also artificially removed and displaced
394
within the landslide, explaining e.g., the anomalous uplift part just downslope of the
395
main scarp close to the Hekkebrug road. Fig. 5D sums up the vertical ground
396
displacements induced by the landslide reactivation and human intervention mainly
397
since February 1995.
398
399
The volume change of the landslide can be estimated from the comparison of the DTMs
400
(Weber and Herrmann, 2000; van Westen and Getahum, 2003; Baldi et al., 2005;
401
Hapke, 2005; Brückl et al., 2006; Chen et al., 2006). According to this comparison, and
402
removing road improvements from the computation, a loss of volume of ~5 000±1 000
403
m³ was recorded between 1952 and 1973. Between 1973 and 1996, ~ 37 000±5 000 m³
404
were lost in the upper part of the landslide while ~ 38 000±6 000 m³ were added
18
405
downslope. This slightly positive balance is likely to result from soil decompaction
406
during sliding. However, the calculated volume expansion is probably underestimated
407
because it also includes collapse and/or erosion of the slipped mass prior the
408
reactivation (1973-1995). From 1996 to 2002, a loss of volume of ~35 000±6 000 m³
409
was detected. This loss of volume is high in comparison with the loss calculated
410
between 1952 and 1973. Insofar as it would not result from problem of vertical
411
inaccuracy induced by the intervention of another operator for the extraction of the
412
LIDAR data, this difference resulted mainly from natural and artificial compaction of
413
the recently displaced soil, and an unknown but possibly high amount of erosion
414
induced by the improvements works linked partly to the construction of the two
415
longitudinal drains (Figs 5C and E).
416
417
5.2. Horizontal displacements
418
419
Several authors computed landslide horizontal displacements from multi-temporal
420
ortho-rectified aerial photographs (e.g. Kääb, 2002; Casson et al., 2003). However, this
421
technique was not applied here due to the scarcity of homologous points between
422
different epochs within the landslide (Dewitte, 2006). The only way to compute
423
horizontal displacements was then to compare the 3D positions of visually well-
424
identifiable geomorphological features (Powers et al., 1996; Gentili et al., 2002).
425
426
We measured horizontal ground displacements both at the main scarp and within the
427
landslide (Fig. 6). As mentioned above, the main scarp is located at the border of a
428
plateau, and its retreat may be compared to a horizontal displacement, locally reaching
19
429
20 m between February 1995 and April 2006 (Fig. 6A). Obviously, this retreat did not
430
result from a single reactivation affecting the whole scarp, but rather from successive
431
collapses like that observed in 2002 (Fig. 5F).
432
433
Fig. 6
434
435
The position of walls located between houses 2 and 3 was also measured in 1973 and
436
1996 (Fig. 6). Whereas the walls next to house 3 were not affected by the reactivation,
437
the displacements of the walls in front of house 2 clearly testify to the lateral movement
438
of this area (Fig. 6B + pictures). These movements were generally larger along the road
439
(Fig. 6B), i.e. closer to the centre of the moving mass where the shear strength is lower
440
(Powers et al., 2006). Road displacements within the transfer area also bear witness to
441
lateral motion (Fig. 6), but no well-identifiable point allowed an accurate estimate of
442
movement.
443
444
Topographic profiles are also very illustrative of the combination of vertical and
445
horizontal displacements (Fig. 7). Fig. 7A shows that the main scarp retreat, amounting
446
to 14 m in this profile, is clearly associated with the zone of the highest subsidence at
447
the landslide head. According to the conceptual model developed by Casson et al.
448
(2005) for rotational landslides, the distribution of the topographic variations between
449
1973 and 1996 corresponds to a large rotation angle dividing the landslide in three areas
450
(depletion-transfer-accumulation). According to this model, the distribution of elevation
451
changes within the profile 1973 – 1996, the shape of the visible part of the surface of
452
rupture at the landslide head, and the alternation of subhorizontal clayey sand and clay
20
453
layers suggest that the surface of rupture of the February 1995 reactivation is situated at
454
a depth of ~20 m (Fig. 7A). It is in contact with the Aalbeke clays and stretches on a
455
length of ~220 m (between 100 and 320 m from the origin of the profile). The uplift in
456
the zone of accumulation is not a true vertical movement of the ground, but mainly the
457
effect of the downslope lateral displacement of the slipped mass along the curved
458
rupture surface (Fig. 7A). Due to the movement along this curved surface of rupture, the
459
horizontal component of this displacement is therefore lower than the horizontal
460
distance between the two profiles (10 m) that would result from a movement on a
461
horizontal sliding plane. Likewise, the transfer area where no vertical displacement of
462
the surface is observed between the collapsed and uplifted parts of the landslide
463
corresponds to a zone of mainly horizontal but smaller motion. This model agrees with
464
the values of lateral displacement measured at walls (Fig. 6B). The small uplifts located
465
downslope of the intersection between the lower part of the surface of rupture and the
466
original ground surface; i.e. the toe of the surface of rupture (Fig. 7A), correspond very
467
well with several small earth flows which followed the main reactivation (Van Den
468
Eeckhaut et al., 2007a).
469
470
Fig. 7
471
472
An interesting issue is to know whether this deep reactivation affected the pre-existing
473
rupture surface of the landslide or developed into the landslide debris. Electric
474
resistivity profiles measurements carried out across a similar but dormant rotational
475
earth slide by Van Den Eeckhaut et al. (2007b) revealed a surface of rupture at 15 m
476
depth, i.e. pretty similar to what we found for the reactivation. In addition, their surface
21
477
of rupture was also in contact with the sensitive clays of the Aalbeke Member (Fig. 7A).
478
Dewitte and Demoulin (2005) however estimated a depth of the surface of rupture of
479
landslide 1 of 40 m, which seems somewhat exaggerated (Fig. 7A). Actually, their
480
measurements relied on the hypothesis that the landslide was a single rotational slide.
481
Yet, this landslide, as many others of the Flemish Ardennes and in particular that
482
studied by Van Den Eeckhaut et al. (2007b), is more probably a multiple rotational
483
slide. Therefore, a part of the ancient surface of rupture of the landslide was probably
484
re-used by the February 1995 reactivation and, consequently, both surfaces of rupture
485
could be very similar. Supposing a depth of the surface of rupture Dr = 20 m, and taking
486
the length and width of the rupture surface as respectively Lr = 266 m and Wr = 226 m
487
(Dewitte and Demoulin, 2005), the initial volume of material before the landslide
488
moved, Volr can be computed as (Cruden and Varnes, 1996):
489
490
1
Volr = πD r ×Wr ×L r
6
(1)
491
492
A volume of ~625 000 m³ is computed, i.e. half the volume computed by Dewitte and
493
Demoulin (2005).
494
495
The 1996-2002 comparison of the topographic profiles shows a succession of positive
496
and negative vertical variations of topography (Fig. 7B). Until 2000 and the
497
construction of the two drainage trenches, several flow-like reactivations affected the
498
zone of accumulation (Van Den Eeckhaut et al., 2007a). Starting from the top, the first
499
subsidence/uplift succession is due to the reactivation of the main scarp in March 2002
500
(e.g. Fig. 5.F). According to the model of Casson et al. (2005) this profile corresponds
22
501
to a weak rotation angle. Therefore, this reactivation did not probably affect the
502
landslide as deeply as the reactivation of February 1995, so that compaction dominated
503
in the largest part of the zone of accumulation. Some of the topographic variations in
504
this area were also the result of anthropogenic material displacements.
505
506
6. Conclusion
507
508
This research aimed at describing and understanding the kinematics of small landslide
509
reactivation in the Flemish Ardennes through the use of multi-temporal DTMs. It
510
demonstrated the effectiveness of multi-temporal DTMs, constructed from aerial
511
stereophotogrammetry using standard photographs and airborne LIDAR data, in
512
retrieving small ground displacements over a period of several decades, and discussing
513
the superficial and inner landslide kinematics.
514
515
An example of ground motion analysis has been given for one representative landslide,
516
showing the spatial and temporal variability of the displacements. The observed height
517
changes highlight the rotational pattern associated with a reactivation in February 1995,
518
and the later compaction as well. Estimates of the horizontal displacements corroborate
519
the analysis.
520
521
Acknowledgements
522
23
523
We gratefully thank V. Havart, D. Weverbergh, and T. Steux for their technical support
524
during the photogrammetric analysis. We also thank the editor and two anonymous
525
reviewers for their relevant comments that helped in improving the manuscript.
526
527
References
528
529
Antonello, G., Casagli, N., Farina, P., Leva, D., Nico, G., Sieber, A.J., Tarchi, D., 2004.
530
Ground-based SAR interferometry for monitoring mass movements. Landslides,
531
1(1): 21-28.
532
Baldi, P., Fabris, M., Marsella, M., Monticelli, R., 2005. Monitoring the morphological
533
evolution of the Sciara del Fuoco during the 2002-2003 Stromboli eruption
534
using multi-temporal photogrammetry. ISPRS Journal of Photogrammetry and
535
Remote Sensing, 59(4): 199-211.
536
Berthier, E., Vadon, H., Baratoux, D., Arnaud, Y., Vincent, C., Feigl, K.L., Rémy, F.,
537
Legrésy, B., 2005. Surface motion of mountain glaciers derived from satellite
538
optical imagery. Remote Sensing of Environment, 95(1): 14-28.
539
540
541
Borradaile, G., 2003. Statistics of earth science data: their distribution in time, space,
and orientation. Springer-Verlag, Berlin, 351 pp.
Brückl, E., Brunner, F.K., Kraus, K., 2006. Kinematics of a deep-seated landslide
542
derived from photogrammetric, GPS and geophysical data. Engineering
543
Geology, 88(3-4): 149-159.
544
Buiten, H.J., van Putten, B., 1997. Quality assessment of remote sensing image
545
registration - Analysis and testing of control point residuals. ISPRS Journal of
546
Photogrammetry and Remote Sensing, 52(2): 57-73.
24
547
Casson, B., Delacourt, C., Allemand, P., 2005. Contribution of multi-temporal remote
548
sensing images to characterize landslide slip surface - Application to the La
549
Clapiere landslide France). Natural Hazards and Earth System Sciences, 5(3):
550
425-437.
551
Casson, B., Delacourt, C., Baratoux, D., Allemand, P., 2003. Seventeen years of the "La
552
Clapiere" landslide evolution analysed from ortho-rectified aerial photographs.
553
Engineering Geology, 68(1-2): 123-139.
554
Chandler, J., 1999. Effective application of automated digital photogrammetry for
555
geomorphological research. Earth Surface Processes and Landforms, 24(1): 51-
556
63.
557
Chaplot, V., Darboux, F., Bourennane, H., Leguedois, S., Silvera, N., Phachomphon,
558
K., 2006. Accuracy of interpolation techniques for the derivation of digital
559
elevation models in relation to landform types and data density.
560
Geomorphology, 77(1-2): 126-141.
561
Chen, R.F., Chang, K.J., Angelier, J., Chan, Y.C., Deffontaines, B., Lee, C.T., Lin,
562
M.L., 2006. Topographical changes revealed by high-resolution airborne LiDAR
563
data: The 1999 Tsaoling landslide induced by the Chi-Chi earthquake.
564
Engineering Geology, 88(3-4): 160-172.
565
Coe, J.A., Ellis, W.L., Godt, J.W., Savage, W.Z., Savage, J.E., Michael, J.A., Kibler,
566
J.D., Powers, P.S., Lidke, D.J., Debray, S., 2003. Seasonal movement of the
567
Slumgullion landslide determined from Global Positioning System surveys and
568
field instrumentation, July 1998-March 2002. Engineering Geology, 68(1-2): 67-
569
101.
25
570
Cruden, D.M. and Varnes, D.J., 1996. Landslide types and processes. In: A.K. Turner
571
and R.L. Schuster (Editors), Landslides: investigation and mitigation.
572
Transportation Research Board, Special Report 247, National Research Council.
573
National Academy Press, Washington D.C., pp. 36-75.
574
DEM of Flanders, 2007. Digital elevation model of Flanders (obtained by
575
laserscanning). http://www.gisvlaanderen.be/gis/diensten/geo-
576
vlaanderen/?catid=8.
577
Demoulin, A., 2006. Monitoring and mapping landslide displacements: a combined
578
DGPS-stereophotogrammetric approach for detailed short- and long-term rate
579
estimates. Terra Nova, 18(4): 290-298.
580
Dewitte, O., 2006. Kinematics of landslides in the Oudenaarde area and prediction of
581
their reactivation: a probabilistic approach, Ph.D. Thesis, University of Liège,
582
Liège, 213 pp.
583
Dewitte, O., Chung, C.-J., Demoulin, A., 2006. Reactivation hazard mapping for
584
ancient landslides in West Belgium. Natural Hazards and Earth System
585
Sciences, 6(4): 653-662.
586
Dewitte, O., Demoulin, A., 2005. Morphometry and kinematics of landslides inferred
587
from precise DTMs in West Belgium. Natural Hazards and Earth System
588
Sciences, 5(2): 259-265.
589
Dikau, R., Brunsden, D., L., S., Isben, M.-L. (Editors), 1996. Landslide recognition.
590
Identification, movement and causes. International Association of
591
Geomrphologists, 5. Wiley, Chichester, 274 pp.
26
592
Fabris, M., Pesci, A., 2005. Automated DEM extraction in digital aerial
593
photogrammetry: precisions and validation for mass movement monitoring.
594
Annals of Geophysics, 48(6): 973-988.
595
Gentili, G., Giusti, E., Pizzaferri, G., 2002. Photogrammetric techniques for the
596
investigation of the Cornoglio landslide. In: R.J. Allison (Editor), Applied
597
geomorphology: theory and practice. International Association of
598
Geomorphologist, Publication No. 10. Wiley, Chichester, pp. 39-48.
599
Glenn, N.F., Streutker, D.R., Chadwick, D.J., Thackray, G.D., Dorsch, S.J., 2006.
600
Analysis of LiDAR-derived topographic information for characterizing and
601
differentiating landslide morphology and activity. Geomorphology, 73(1-2):
602
131-148.
603
604
605
Goovaerts, P., 1997. Geostatistics for natural resources evaluation. Applied geostatistics
series. Oxford University Press, Oxford, 483 pp.
Hancock, G., Willgoose, G., 2001. The production of digital elevation models for
606
experimental model landscapes. Earth Surface Processes and Landforms, 26(5):
607
475-490.
608
Hapke, C.J., 2005. Estimation of regional material yield from coastal landslides based
609
on historical digital terrain modelling. Earth Surface Processes and Landforms,
610
30(6): 679-697.
611
612
Higgitt, D.L., Warburton, J., 1999. Applications of differential GPS in upland fluvial
geomorphology. Geomorphology, 29(1-2): 121-134.
613
Jacobs, P., De Ceukelaire, M., De Breuck, W., De Moor, G., 1999. Toelichtingen bij de
614
geologische kaart van België - Wlaams Gewest, Kaartblad (29) Kortrijk
27
615
1:50.000. Ministerie van Economische Zaken and Ministerie van de Vlaamse
616
Gemeenschap, 68 pp.
617
Kääb, A., 2002. Monitoring high-mountain terrain deformation from repeated air- and
618
spaceborne optical data: examples using digital aerial imagery and ASTER data.
619
ISPRS Journal of Photogrammetry and Remote Sensing, 57(1-2): 39-52.
620
621
622
623
624
625
626
Kasser, M., Egels, Y., 2001. Photogrammétrie numérique. Collection ENSG - IGN.
Hermes, Paris, 379 pp.
Keefer, D.K., 1984. Landslides caused by earthquakes. Geological Society of America
Bulletin, 95(4): 406-421.
Kraus, K., Waldhäusl, P., 1994. Photogrammetry: fundamentals and standard processes,
1. Dümmler Verlag, Bonn, 397 pp.
Lantuit, H., Pollard, W.H., 2005. Temporal stereophotogrammetric analysis of
627
retrogressive thaw slumps on Herschel Island, Yukon territory. Natural Hazards
628
and Earth System Sciences, 5(3): 413-423.
629
630
631
LH Systems, 1999. Digital Photogrammetric Workstation: SOCET SET Version 4.2
User's Manual. Marconi Integrated Systems Inc., 802 pp.
Malet, J.P., Maquaire, O., Calais, E., 2002. The use of Global Positioning System
632
techniques for the continuous monitoring of landslides: application to the Super-
633
Sauze earthflow (Alpes-de-Haute-Provence, France). Geomorphology, 43(1-2):
634
33-54.
635
Mantovani, F., Soeters, R., VanWesten, C.J., 1996. Remote sensing techniques for
636
landslide studies and hazard zonation in Europe. Geomorphology, 15(3-4): 213-
637
225.
28
638
McKean, J., Roering, J., 2004. Objective landslide detection and surface morphology
639
mapping using high-resolution airborne laser altimetry. Geomorphology, 57(3-
640
4): 331-351.
641
Metternicht, G., Hurni, L., Gogu, R., 2005. Remote sensing of landslides: An analysis
642
of the potential contribution to geo-spatial systems for hazard assessment in
643
mountainous environments. Remote Sensing of Environment, 98(2-3): 284-303.
644
645
646
Mikhail, E.M., Bethel, J.S., McGlone, J.C., 2001. Introduction to modern
photogrammetry. Wiley, New York, 479 pp.
Nico, G., Rutigliano, P., Benedetto, C., Vespe, R., 2005. Terrain modelling by
647
kinematical GPS survey. Natural Hazards and Earth System Sciences, 5(2): 293-
648
299.
649
650
651
Oka, N., 1998. Application of photogrammetry to the field observation of failed slopes.
Engineering Geology, 50(1-2): 85-100.
Powers, P.S., Chiarle, M., Savage, W.Z., 1996. A digital photogrammetric method for
652
measuring horizontal surficial movements on the Slumgullion earthflow,
653
Hinsdale county, Colorado. Computers & Geosciences, 22(6): 651-663.
654
Siska, P.P., Goovaerts, P., Hung, I.K., Bryant, V.M., 2005. Predicting ordinary kriging
655
errors caused by surface roughness and dissectivity. Earth Surface Processes and
656
Landforms, 30(5): 601-612.
657
Squarzoni, C., Delacourt, C., Allemand, P., 2005. Differential single-frequency GPS
658
monitoring of the La Valette landslide (French Alps). Engineering Geology,
659
79(3-4): 215-229.
660
661
Tarchi, D., Casagli, N., Fanti, R., Leva, D.D., Luzi, G., Pasuto, A., Pieraccini, M.,
Silvano, S., 2003. Landslide monitoring by using ground-based SAR
29
662
interferometry: an example of application to the Tessina landslide in Italy.
663
Engineering Geology, 68(1-2): 15-30.
664
Van Den Eeckhaut, M., 2006. Spatial and temporal patterns of landslide in hilly regions.
665
The Flemish Ardennes (Belgium), Katholieke Universiteit Leuven, Leuven, 250
666
pp.
667
Van Den Eeckhaut, M., Poesen, J., Dewitte, O., Demoulin, A., De Bo, H., Vanmaercke-
668
Gottigny, M.C., 2007a. Reactivation of old landslides: lessons learned from a
669
case-study in the Flemish Ardennes (Belgium). Soil Use and Management 23(2):
670
200-211.
671
Van Den Eeckhaut, M., Poesen, J., Verstraeten, G., Vanacker, V., Moeyersons, J.,
672
Nyssen, J., van Beek, L.P.H., 2005. The effectiveness of hillshade maps and
673
expert knowledge in mapping old deep-seated landslides. Geomorphology, 67(3-
674
4): 351-363.
675
Van Den Eeckhaut, M., Verstraeten, G., Poesen, J., 2007b. Morphology and internal
676
structure of a dormant landslide in a hilly area: The Collinabos landslide
677
(Belgium). Geomorphology, 89(3-4): 258-273.
678
van Westen, C.J., Getahun, F.L., 2003. Analyzing the evolution of the Tessina landslide
679
using aerial photographs and digital elevation models. Geomorphology, 54(1-2):
680
77-89.
681
Weber, D., Herrmann, A., 2000. Contribution de la photogrammétrie numérique à
682
l'étude spatio-temporelle de versants instables : l'exemple du glissement de
683
Super-Sauze (Alpes-de-Haute-Provence). Bulletin de la Société géologique de
684
France, 171(6): 637-648.
685
30
686
Table captions
687
688
Table 1. Main characteristics of the photograph negatives used for the digital
689
stereophotogrammetry.
690
691
Table 2. Root mean square error (RMSE) of the interior orientation.
692
693
Table 3. Main characteristics of the 3 aerotriangulations of 1996, 1973 and 1952. The
694
unknowns are the orientation parameters of the images and the adjusted XYZ
695
coordinates of the homologous points. The knowns are the XYZ coordinates of the
696
ground control points and the ξ and η image coordinates of the control points and tie
697
points.
698
699
Table 4. Precision and accuracy of the 1996, 1973 and 1952 aerotriangulations.
700
701
Table 5. Accuracy of the DTMs.
702
703
Table 6. Accuracy and confidence intervals of the DTM subtractions.
704
705
706
Figure captions
707
708
Fig. 1. (A) Location map of the Flemish Ardennes and the study area. (B) Location of
709
the two studied hills with the lithological setting, and the boundaries and main scarps of
31
710
the 13 landslides considered in the analysis. The quadrangles DTM I and DTM II
711
delimit the two areas where the DTMs were extracted.
712
713
Fig. 2. Flow chart of the photogrammetric operations carried out to capture the 3D
714
morphology of the landslides.
715
716
Fig. 3. Image residuals of the ground points used in the three aerotriangulations (1996,
717
1973 and 1952). The ground points (tie points and ground control points) are placed at
718
their initial location. The residuals indicate thus the direction and the range of their
719
displacements towards their adjusted locations. For example, a point whose residual is
720
oriented towards the east is a point which has undergone during its adjustment a
721
displacement towards the east. A ground point with two residuals is a point measured in
722
two images. The image residuals are exaggerated 2000 × in the block of 1996 and 500 ×
723
in the blocks of 1973 and 1952.
724
725
Fig. 4. XYZ residuals of the check points used in the three aerotriangulations (1996,
726
1973 and 1952).The check points are placed at their adjusted location. The residuals
727
indicate the location error of these points compared to the ground coordinates measured
728
by GPS. For example, a negative residual in Z (a Z residual is positive towards the north
729
and negative towards the south) means that the adjusted point of the model is at an
730
altitude higher than the GPS-defined altitude of the point; a positive residual in XY (a
731
residual in XY is positive towards the east and negative towards the west) indicates that
732
the adjusted point of the model is at the west of the point measured by GPS. The XYZ
733
residuals are exaggerated 2000 × in the 3 blocks.
32
734
735
Fig. 5. Maps of vertical ground displacements within landslide 1 inferred from the
736
comparison between the DTMs. (A): from 1952 to 1973; (B): from 1973 to 1996; (C):
737
from 1996 to 2002; (D): from 1952 to 2002. In view (A) and (B), the underlying
738
contour lines describe respectively the 1973 and 1996 topographies whereas the 2002
739
topography is shown in views (C) and (D). Only the vertical movements larger than 1 σ
740
(i.e. 0.75 m) uncertainty are represented. (E) View towards the west from the top of the
741
main scarp of two 2-m-deep drainage trenches dug in 2000 (Picture taken in February
742
2003). A big part of the young trees growing within the landslide are popular trees
743
planted in place of large poplar tress cut in September 1995 after the main reactivation.
744
(F) Main scarp reactivation occurred in March 2002 showing clearly the listric shape of
745
the slip surface (Picture taken in May 2002).
746
747
Fig. 6. Horizontal displacements measured within landslide 1 between 1973 and 1996.
748
(A) Global view. (B) Focussed view on wall displacement close to house 2. The
749
collapsed and uplifted parts between 1973 and 1996 are also represented. The three
750
views, towards the south, witness to traces of horizontal displacements (April 2004).
751
View A: picture taken from the Hekkebrug road upslope house 2. The backward
752
toppling of the trees towards the landslide head is due to lateral displacement of the
753
terrain. View B: picture taken from the Hekkebrug road in front of house 2. The main
754
entry, which faced the house, has moved several meters downslope since the February
755
1995 reactivation. View C: picture taken on the road crossing the toe of the landslide.
756
The backward toppling of the trees towards the landslide head is also due to lateral
757
displacement of the terrain.
33
758
759
Fig. 7. DTM cross-sections of landslide 1 compared with the vertical displacements. (A)
760
Ground displacements between 1973 and 1996. (B) Ground displacements between
761
1996 and 2002. Lateral motion of the main scarps and ridges are represented with
762
horizontal arrows, vertical motions with vertical arrows, and rotational motions with
763
curved arrows. The curve associated with the second Y-axis (on the right) corresponds
764
to the vertical displacements extracted from the differential DTMs. The empty area
765
between the two horizontal dashed lines delimits the ± 1σ confidence band resulting
766
from the DTM subtraction. Only the vertical movements in excess of ± 1 σ (i.e. 0.75 m)
767
uncertainty are represented, which delimits the collapsed and uplifted parts. For each
768
curve, the maximum and minimum vertical displacements are noted. The profiles are
769
located on Fig. 6.
770
771
34
772
Table 1
773
774
775
35
776
Table 2
777
778
779
780
36
781
Table 3
782
783
784
785
37
786
Table 4
787
788
789
38
790
Table 5
791
792
793
39
794
Table 6
795
796
797
40
798
Figure 1
799
800
801
802
41
803
Figure 2
804
805
806
807
42
808
Figure 3
809
810
811
43
812
Figure 4
813
814
815
816
44
817
Figure 5
818
819
820
45
821
Figure 6
822
823
824
825
46
826
Figure 7
827
828
829
47